Skill or luck? Reading a crypto trading record

We tested this directly. Simulated traders drew 90 days at a time from the day shapes of real records on Sharpe, with the record's own edge over Bitcoin taken out, and for a skilled trader an edge of a known size put back. Every draw was scored by the same engine that ranks the board.

A trader with no edge at all reached Diamond or above in one window 23.5% of the time. A trader with a true annual information ratio of 2 against Bitcoin, which is strong, reached it 59.8% of the time. The skilled trader is more likely to get there, but a quarter at the top is something an unskilled trader manages about one time in four, and a board of many traders will always have some of them near the top.

Two records can finish the same distance ahead of Bitcoin and be very different evidence. One got there by beating Bitcoin by a little on most days. The other spent most of its days behind and made the whole lead on two or three. The second record's result depends on those few days, and a slightly different quarter without them would read as a loss.

Statistics puts a number on this. The information ratio, a record's average daily lead over Bitcoin divided by how much that lead swings, has a standard error, and the error grows when a few large days dominate. Sharpe discounts every record by half of it, so a lead counts for less when it rests on little. Over 90 days the discount for a record of ordinary shape is about 1.0 of annual ratio; over a year it would be about 0.5.

Beside every rank on the leaderboard, Sharpe shows Confidence: the probability that the record's true excess over Bitcoin is above zero, given what its days show. It is the probabilistic Sharpe ratio of Bailey and López de Prado (2012), and Sharpe reports it in five bands. Established means 95% or more, Likely 80%, Tentative 60% and Unproven 40%. Below 40% the evidence points the other way, and the band is Trailing.

The same paper gives the number of days a record of a given shape would need before its lead cleared the usual 95% test. The answer is sobering. For a record of ordinary shape, an annual ratio of about 3.4 establishes itself within 90 days. A ratio of 1.0, a level the standard textbook on active management calls exceptional, takes 991 days.

Luck rarely repeats, which is why a second window helps so much. In the same simulation, Master in both of two consecutive windows was reached by 2.9% of traders with no edge, 10.0% of traders with an annual ratio of 1, and 25.2% of traders with a ratio of 2. Sharpe marks a Diamond or Master record whose previous window measured the same tier, so a reader can tell a repeat from a first appearance.

We also split real records in time and asked whether the score of the earlier half ordered records by how the later half did against Bitcoin. It did, mostly at the bottom. A score below the Bitcoin line said a good deal about what came next. At the top the signal was weaker: of the records whose earlier half measured Master, 31% beat Bitcoin over the later half, against 19% of all records. That is better than the board as a whole, and far from a promise.

A rank describes the 90 days it measures. Read it with its Confidence, and with whether it has held for two windows, and it tells you a good deal about how much of a record is skill. The full tests are in the methodology.